There’s a lot of conversation at the moment about AI agents and what they mean for traditional business systems. The question I get asked most is a simple one: will AI replace your ERP?
Will businesses still need integration platforms? Could an AI agent simply connect to Shopify, an ERP, a warehouse system and a bank, and manage everything itself?
I don’t think that’s where we’re heading.
I think AI is going to make these systems significantly more powerful, but the underlying technology will still have an important job to do.
The easiest way I think about it is this:
Your business systems hold the data. Integration moves the data. AI understands the data.
And increasingly, AI will also be able to act on it.
Integration still has a job to do
If you run an eCommerce business and want an order placed in Shopify to automatically appear in your ERP system, you still need something to move that data reliably from one system to another.
That’s what integration is good at. It removes double entry and handles predictable, repeatable processes incredibly well.
The interesting question is what happens when the process is not predictable. That’s where I think AI becomes particularly valuable.
Take a relatively simple example. A customer places an order online but enters an incorrect or incomplete delivery address. The integration attempts to create the order in the ERP, but the validation fails.
Traditionally, the process stops. Someone has to investigate the error, go back into Shopify, correct the address, and then allow the integration to reprocess the order.
The integration has done exactly what it was designed to do. This is the exception that requires human judgement.
Now imagine adding AI to that process. Instead of simply reporting that the order has failed, AI could analyse the address, determine the likely correction, update it, and allow the order to continue.
That changes the role of automation considerably.
I wouldn’t give AI complete control on day one
There’s obviously risk involved. What if AI changes an address incorrectly and a valuable order ends up delivered to the wrong place?
My approach would be to introduce AI gradually.
For the first month, AI could identify the problem and recommend the correction, but a person would still approve the change. You can then compare what the AI is doing with what your team would have done. Once you have enough confidence in its decisions, straightforward cases can become fully automated.
I’d also want a dashboard showing every exception, what the AI changed, and why. That creates an audit trail and keeps the process visible.
The journey therefore becomes:
Human fixes the problem → AI recommends the fix → human validates AI → AI fixes the problem → human monitors AI.
That, to me, is a far more realistic way of introducing autonomous AI into business processes than simply switching it on and hoping for the best.
Reconciliation is another obvious opportunity
Another process we regularly come across is reconciliation.
Imagine trying to reconcile payments across a Shopify report, a bank statement and an ERP system. Automation can collect and move the information, but somebody may still need to compare records, investigate discrepancies, and decide why something doesn’t match.
That’s exactly the kind of activity where I think AI can add another layer. Rather than simply presenting somebody with three sets of data, AI can analyse them, identify discrepancies, and potentially explain why they occurred.
The integration moves the information. The AI interprets it.
But don’t start with AI
If a business came to me tomorrow and said, “We want to use AI, but we don’t know where to start,” I wouldn’t begin by looking for places to install AI.
I’d start by looking at the business processes.
Where are people manually checking information? Where are they correcting errors? Where are they reconciling data between different systems? Where are people repeatedly making the same decisions?
Then I’d identify which of those processes consumes the most time, and start there.
There’s another important step before automating anything:
Should this process actually work this way in the first place?
This comes up in almost every integration project we work on. A customer may ask us to automate an existing process, but once we start mapping it, we realise there’s an opportunity to simplify the process itself.
B2C customer accounts in an ERP are a good example. Historically, businesses sometimes wanted individual customer accounts created for every online customer. Increasingly, businesses use a single B2C account within the ERP to hold those transactions.
We see similar conversations around special pricing, price lists, and particularly refunds. Refund processes often sound incredibly complicated when they’re first explained. Once we work through what actually needs to happen, there’s frequently an opportunity to simplify the process before we automate it.
There’s little value in using great technology to automate a bad process.
The biggest return isn’t always the salary saving
Suppose three people collectively spend 30 hours every week checking information, fixing errors, and reconciling data.
If integration and AI reduce that to two or three hours of oversight, calculating the financial benefit is relatively straightforward. Thirty hours multiplied by an employee cost gives you a tangible saving.
But for a growing business, I don’t think that’s necessarily the biggest benefit.
Data accuracy and speed can be much more valuable.
If an order moves from your website into your ERP faster, through your warehouse faster, and ultimately reaches your customer faster, you’ve improved something far more important than an internal administration process. You’ve improved the customer journey.
Amazon is probably the easiest example. People buy from Amazon for many reasons, including the huge range of products. But speed and convenience are a massive part of the proposition. Order something today and, depending on the product and location, it could arrive the same day or the next.
Customers value that experience so much that millions of them pay a subscription for Prime. That demonstrates something important.
Operational efficiency eventually becomes customer experience. And customer experience eventually becomes commercial performance.
A customer who consistently receives the correct order quickly is more likely to come back, more likely to leave a positive review, and more likely to recommend the company. Those reviews and recommendations make marketing more effective, which helps attract more customers.
So when calculating the return from automation, I’d absolutely look at hours and salary savings. But I’d also ask: How much faster can we process an order? How many errors can we prevent? How quickly can customer service understand what happened when something goes wrong? How many more orders can the business process without increasing operational overhead at the same rate?
And ultimately, what does that do to the customer experience?
Most businesses don’t start there
From my experience, around 9 out of 10 businesses we speak to initially approach automation because they want to remove a manual process.
“We’re manually entering Shopify orders into our ERP.” “We’re copying information between these two systems.” “Someone spends hours every week doing this.”
Very rarely does somebody come into the first conversation and say: “I want to automate this because I want to improve our customer journey.”
That tends to come later. Once the first integration is running and the business sees the impact, the conversation changes. They start asking what else can be improved.
That’s often when automation stops being viewed purely as an IT project or a way of reducing administration, and starts becoming part of how the business grows.
So where does AI fit?
I think the technology stack over the next few years will become increasingly clear.
Your ERP, eCommerce platform, WMS and other applications will continue to be your systems of record. Integration platforms will continue to be the plumbing, reliably moving information between those systems. AI will increasingly become the intelligence layer sitting across those processes — analysing information, validating it, identifying exceptions, correcting problems, reconciling records, and making certain decisions. And people will increasingly become the oversight layer, focusing on unusual situations, strategic decisions, and work where human judgement genuinely adds value.
I don’t believe AI will replace ERP systems or traditional integration technology. I think it will make both significantly more powerful.
And for businesses wondering where to start, I wouldn’t start with the question, “Where can we use AI?” I’d start with: “Where are we wasting the most time, where are errors affecting our customers, and what’s stopping this process from moving faster?”
Solve that problem first. Then decide whether the answer is better integration, better automation, AI — or, increasingly, a combination of all three.
If you’re working through where AI and automation fit in your own systems, that’s exactly the kind of conversation we have with clients every week. Get in touch and we’ll talk it through.